Visual moving object tracking via sparse representation based trackers: A comparative study

Motion tracking is one of the richest research fields in computer vision. Indeed, numerous algorithms have been implemented for object tracking. In this paper we briefly present the principles of three recent methods treating the motion tracking: the discriminative sparse similarity map (DSS map), the probability continuous outlier model (PCOM) and the L2 regularized least square (L2-RLS). And then we evaluate them quantitatively and qualitatively by testing them on nine image sequences which including various challenging factors. In order to achieve that, two most popular criterions: the center location error and the overlap rate are computed for each method. The most effective tracker is the one that has the greatest overlap rate and the smallest center error.